arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2108.01772math.OCcs.ITcs.LGeess.SPmath.ITstat.ML
收藏

低秩矩阵优化中非凸因子分解与流形公式几乎等价

Nonconvex Factorization and Manifold Formulations are Almost Equivalent in Low-rank Matrix Optimization

University of Chicago(芝加哥大学);Fudan University(复旦大学);Duke University(杜克大学)

Yuetian Luo, Xudong Li, Anru R. Zhang

AI总结:

该研究建立低秩矩阵优化中流形与因子分解公式在平稳点、鞍点等上的几何等价性及谱夹心关系,可迁移几何性质、解决文献遗留问题并强化相关应用结论。

摘要:

本文研究了低秩半正定(PSD)及一般矩阵优化中被广泛研究的流形公式与因子分解公式的几何景观关联。我们在一阶平稳点(FOSPs)处,建立了黎曼海森矩阵与欧几里得海森矩阵谱之间的夹心关系。由此,我们得到了两种公式在一阶平稳点、二阶平稳点(SOSPs)和严格鞍点集合上的等价性。此外,我们证明该夹心关系可用于将更多定量几何性质从一种公式迁移到另一种。本文还讨论了PSD情形与一般情形下景观关联的异同点。据我们所知,这是首个针对秩约束的流形公式与因子分解公式之间的几何景观关联,它为文献中观测到的低秩矩阵优化里因子分解方法与流形方法经验性能相似的现象提供了几何层面的解释。在一般低秩矩阵优化中,本文还给出了两种因子分解公式(无正则与有正则)的景观关联。通过应用这些几何景观关联,尤其是夹心关系,我们得以解决文献中悬而未决的问题,并在相位恢复的几何分析、良态低秩矩阵优化,以及机器学习与信号处理中因子分解里正则化的作用等应用场景中得到了更强的结果。

英文摘要:

In this paper, we consider the geometric landscape connection of the widely studied manifold and factorization formulations in low-rank positive semidefinite (PSD) and general matrix optimization. We establish a sandwich relation on the spectrum of Riemannian and Euclidean Hessians at first-order stationary points (FOSPs). As a result of that, we obtain an equivalence on the set of FOSPs, second-order stationary points (SOSPs) and strict saddles between the manifold and the factorization formulations. In addition, we show the sandwich relation can be used to transfer more quantitative geometric properties from one formulation to another. Similarities and differences in the landscape connection under the PSD case and the general case are discussed. To the best of our knowledge, this is the first geometric landscape connection between the manifold and the factorization formulations for handling rank constraints, and it provides a geometric explanation for the similar empirical performance of factorization and manifold approaches in low-rank matrix optimization observed in the literature. In the general low-rank matrix optimization, the landscape connection of two factorization formulations (unregularized and regularized ones) is also provided. By applying these geometric landscape connections, in particular, the sandwich relation, we are able to solve unanswered questions in literature and establish stronger results in the applications on geometric analysis of phase retrieval, well-conditioned low-rank matrix optimization, and the role of regularization in factorization arising from machine learning and signal processing.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑